Foundations
- LLM, rules or plain search
- Prompting, RAG or fine-tuning
- Retrieval quality
DataKern Certification
DataKern Certified AI Engineer
LLMs, retrieval, agents, evaluation and running AI in production: what to build, and what to leave out.
Launch offerFree to try out. Practice and the exam cost nothing while we launch. Register, try a case, and see what you can decide.
Associate level
DataKern Certified AI Engineer (Associate)
- 45 questions
- 9 cases
- 90 minutes
- Open-book
Open to any registered student. Every case asks five questions, one for each of the five moves.
What is tested
Building
- Workflow or agent
- Tools and guardrails
- Evaluation
Operating
- Model choice for cost, latency and quality
- Privacy and rules
- Tracing and monitoring
Professional level
DataKern Certified AI Engineer (Professional)
- 30 questions
- 6 cases
- 90 minutes
- Open-book
Opens after you pass Associate in the same track. Every case asks five questions, one for each of the five moves.
What is tested
Design
- Single agent or multi-agent
- Build or buy an AI capability
Operate
- Evaluation strategy for a customer-facing assistant
- Cost governance for LLM workloads
Govern and change
- Safety incident response
- Rollout and rollback
What every case looks like
Each case is a short story: an email, sometimes chats, logs or tables, with real limits like team size, budget, deadline and data rules. After it come five questions, always in this order:
- Spot the problem. Read a real situation and find what actually matters, using the numbers in front of you.
- Pick the solution. Pick the best option, and say why. The best choice depends on the numbers.
- Ask the AI. Write a request that an AI assistant can really use: the facts, the limits and the rules.
- Check the plan. An AI assistant suggests a plan. Keep the good parts, fix the weak ones, and reject the risky ones.
- Handle the change. Something changes in the middle of the work. Decide what the plan now needs.
The exam is for you. It is open-book and unproctored, so you can check your notes and the documentation. Please do not ask an AI to answer the questions: the result only helps you if the answers are yours. This is DataKern's own credential, and the badge states the conditions: open-book, unproctored, 90 minutes. Read the rules.